Page 79 of 136
It is important to note that the data do not contain details by port of loading number of sheep loaded and deaths. They only contain the total sheep on the ship and an indication of which ports sheep
were loaded from.
Data for all voyages containing sheep from 2006 to 2011 were compiled. Data relating to voyages with very small consignments of sheep, that were loaded at ports other than Portland, Adelaide or
Fremantle, or that were going to ports other than the Middle East, were removed.
The resulting dataset contained one row per export voyage and included a total of 277 voyages over the 6 year period.
Port of loading was coded as 1 if any sheep were loaded at Portland regardless of whether additional sheep were loaded at other ports, 2 if sheep were loaded at Adelaide including voyages
where sheep were loaded at Adelaide and Fremantle, and 3 if sheep were loaded at Fremantle only.
Voyage duration was recorded in days. Month of loading was coded as 1 to 12 and year was coded as 2006 to 2011.
A numeric variable was used to code for exporter with those exporters associated with very few voyages aggregated into a single category.
6.1 Voyage mortality rate
A negative binomial statistical model was then applied to these data using the count of deaths as the outcome and the total number of sheep loaded onto the ship as an offset or estimate of animal time
at risk. The model included fixed effects coding for exporter, port and month and year of loading, and duration of voyage. The model included an interaction between port and month of loading to account
for possible differences between ports.
There was interest in determining whether there might be an interaction between year and month. The statistical modelling was being run with a particular focus on the question of whether there was
a season pattern that may differ for different ports. Figure 6 is a plot of the means derived from the portmonth interaction from the main statistical model. It seemed plausible that this could be
dependent on year. For example, if there was one year that was markedly different to all other years then this unusual year could influence the pattern of the monthport interaction.
This was investigated in a separate preliminary model that just included year and month and the yearmonth interaction. The results from this model are not shown. There was variability between
years for example the mean mortality rates for any given month between years did vary. However, there was a very clear overall pattern that remained evident in all years – lower mortality in the first
four months of the year, followed by a rise to a peak in August and then a decline towards the end of
Page 80 of 136
the year. This provided confidence that the overall seasonal pattern change by month within any given year was very similar in each year.
This finding provided confidence that the main statistical model could then be fitted as follows: outcome = count of deaths on each voyage adjusted by the total number of sheep on the
voyage mortality rate per voyage expressed as deaths per 100 sheep per voyage fixed effects coding for:
o year 2006, 2007, 2008, 2009, 2010, 2011
o port of loading Portland, Adelaide, Fremantle
o month of loading 1, 2, 3, 4, 5,6 ,7, 8, 9, 10, 11, 12
o duration of voyage days
o exporter
1,2,3,4,5 o
interaction between port and month of loading Notice that a variable coding for exporter was included in the model because we expected that there
would be some variability between exporters in how voyages are managed. Part of this will include unmeasured effects associated with ships that may be operated by only one exporter. This variable
is therefore included in the model to provide adjustment for unmeasured factors at the exporter level and as a result is expected to produce more accurate estimates for other fixed effects in the model.
It is the other fixed effects that are of primary interest and particularly the effects of port and month of loading.
The model was then used to generate estimates of mortality incidence rate for various fixed effects.
Page 81 of 136
Table 6.1: Summary count of voyages listed by port of loading Portland
Adelaide Month Alone +Adel +Freo +
Adel+Freo Alone +Freo +Freo Total
1 9
1 11
21 2 1
5 2
13 21
3 1 8
2 1 13
25 4 1
8 3
10 22
5 4
1 2
10 17
6 8
4 5
17 7
6 5
15 26
8 5
5 1 3
15 29
9 4
1 4 10
19 10 1 1 3
3 21 29
11 1 2 2 2 1 19 27
12 1 6
2 15 24
Total 6 8
68 1 11 26 157 277
The above table demonstrates that on almost all occasions when sheep were loaded at either Portland or Adelaide, the ships then took on additional sheep at Fremantle.
There was a single voyage where sheep were loaded at all three ports.
Page 82 of 136
The following table shows how the port of loading was coded for analysis.
Table 6.2: Count of the number of voyages for each combination of port and month in the dataset Month Portland Adelaide Fremantle Total
1 9 1
11 21
2 6 2
13 21
3 9 3
13 25
4 9 3
10 22
5 5 2
10 17
6 8 4
5 17 7 6
5 15
26 8
10 4 15 29 9 4
5 10
19 10 5
3 21
29 11 5
3 19
27 12 7
2 15
24 Total 83
37 157
277
The main reason for coding port in this way was to try and produce a coding where all months of the year were represented and that provide some measure of effect of each port. When considering an
effect of Portland, it is important to realise that the measure of effect will incorporate a component due to Fremantle and a very small component due to Adelaide but that any differences between
this port level and other levels should be able to be attributed to an effect of Portland.
Page 83 of 136
Figure 6.1: Mean overall mortality rate deaths per 100 sheep per voyage averaged over all exporters, months and ports. Bars represent 95 confidence interval either side of the mean.
There was no difference between the first three years, then there was a significant rise from 2008 to 2009 p=0.02 followed by a decline over the following years to the lowest level in 2011.
The overall mortality rate for 2011 0.71 deaths per 100 sheep was lower than in any of the previous years. The difference was statistically significant when compared to all of the previous
years p0.05 except 2008 when the comparison produced a tendency to significance p=0.06.
Page 84 of 136
Table 6.3: Mortality rate expressed as deaths per 100 sheep for export voyages, by month of year. Estimates derived from a regression model applied to 6 years of voyages. Se=standard error,
CI=confidence interval.
Deaths 95 CI
Month per 100
sheep se
lower upper
1 0.759 0.065 0.631
0.887 2 0.735 0.063
0.613 0.858
3 0.584 0.044 0.497
0.671 4 0.635 0.052
0.532 0.737
5 0.873 0.083 0.711
1.036 6 0.836 0.081
0.678 0.994
7 1.091 0.090 0.914
1.268 8 1.330 0.094
1.145 1.515
9 1.064 0.099 0.870
1.259 10 1.003
0.074 0.857
1.149 11 0.913
0.068 0.779
1.046 12 0.781
0.063 0.657
0.906
Page 85 of 136
Figure 6.2: Plot of mortality rate deaths per 100 sheep per voyage by port of loading and month of loading.
The average line is a plot of the estimates presented in Table 6.3. The lowest mortalities occur in the first 4 months of the year and then rise from May to a peak in August before slowly declining. This is
consistent with previous reports suggesting a general pattern with higher mortalities in the second half of the year.
The pattern appears slightly more variable when plotted by port of loading. It is important to understand what Figure 6.2 is showing. The line labelled Portland is not an estimate
of mortality rate for sheep loaded at Portland. It is an estimate of mortality rate for all those voyages that included sheep loaded at Portland. These voyages often included sheep that were loaded at
Fremantle and less commonly included sheep from Adelaide.
Similarly the line labelled Adelaide is an estimate of mortality rate for those voyages that included sheep loaded at Adelaide. Many of these voyages also included sheep loaded at Fremantle.
The line labelled Fremantle is a direct measure of mortality rate for sheep loaded at Fremantle since these voyages contain only sheep loaded at Fremantle.
Page 86 of 136
The value of this plot lies in the visual comparison of the lines. The lines for Portland and Fremantle both contain sheep loaded at Fremantle. The major difference between these lines is the fact that
the Portland line contains sheep loaded from Portland.
If we can assume that the Fremantle component of a voyage that begins at Portland would include sheep that are not different to those that might be included in a Fremantle only voyage, then the
difference between these two lines is likely to be driven by factors associated with sheep loaded at Portland.
There are a number of useful points that can be made from this graph. For the first four months of the year Jan to April all the lines are very close together and also very
close to the overall average line line produced by aggregating all ports together. This suggests that there is essentially no difference in overall mortality risk between different ports of loading.
For those voyages containing sheep loaded at Portland, there is a rise that becomes apparent in May and that continues to rise to a peak in July and August before declining later in the year. The
general pattern is similar to that seen for Fremantle but with a higher rise and peak.
The pattern for voyages containing sheep that were loaded in Adelaide is more variable. There is a short peak of mortalities in May and then a second peak in October. There are fewer voyages in this
category and only two voyages in May. Neither of these voyages included any reportable mortalities but one voyage had a mortality rate of 1.84 and therefore the rise in the plot line was due to the
effects of a single voyage in that month. This sort of effect needs to be interpreted with caution because there may have been particular circumstances associated with that voyage that may not
reflect general or repeatable trends. Similarly for the rise in October for voyages including Adelaide sheep, there were only three voyages and no voyages with reportable mortalities. The rise was due
to the fact that 2 of these 3 voyages had mortalities at 1.25. As a result it is more difficult to be confident of an overall pattern that may be appropriate for voyages containing sheep loaded in
Adelaide. The data suggest that it may be highly variable because of individual voyage variation.
At the back end of the year from September onwards, the lines converge again and track along a similar path to the overall average line.
Page 87 of 136
Figure 6.3: Mean overall mortality rate deaths per 100 sheep per voyage for each port, averaged over all months and all years. Bars represent 95 confidence interval either side of the mean.
Figure 6.3 shows that mortality rates for Portland voyages were higher than those from the other two ports. However, there was no statistical difference between the three ports when mortality rates
were averaged over all months and all years.
The overall average mortality rate across all ports, years and months was 0.89 deaths 95 confidence interval from 0.84 to 0.94.
Page 88 of 136
Figure 6.4: Plot of average total sheep per voyage arranged by port of loading and month. The top dotted line shows the maximum load for any voyage in each month. The other lines show the average
for each port for each month and the dotted line in the middle of the plot shows the overall average for each month.
The reason for showing Figure 6.4 is to see if the visual display of average total loads might support any hypothesis that part of the elevated mortality rate in a particular month may be influenced by a
tendency to load more sheep onto ships in a particular month. The data shown in Figure 6.4 do not support this hypothesis. In the mid to latter part of the year when mortality rates are at their highest,
the maximal loads per voyage are reduced and the average loads are also reduced, most notably for Fremantle. There is an increase in average load for Portland in September but this does not coincide
with peak mortality rates.
Page 89 of 136
Figure 6.5: Mean mortality rate estimates by year and port. Bars represent 95 confidence intervals.
The above plot was generated from an interaction term involving yearport that was added to the statistical model. It demonstrates that in any given year, there is variability between ports with
respect to average annual mortality rate. Voyages that included sheep loaded from Portland had the highest annual mortality rate in four of the six years and the lowest annual mortality rate in the two
remaining years. It is also useful to note that voyages containing only sheep from Fremantle only had the lowest annual mortality rate in one year and in all remaining years were in the middle.
6.2 Odds of reportable mortality events
